Improving Creditworthiness Prediction Using Preprocessing Stages and Feature Selection

The purpose of this research is to improve the credit worthiness prediction model by applying preprocessing and feature selection stages. At the preprocessing stage, missing value data is removed and standardized with a z-score while the Information Gain and Gini Index are used as feature selection methods. This study uses a credit score classification dataset consisting of 19 features, 7 meta data, and 1 target data. Data were tested with and without preprocessing stages, as well as using feature selection using the Information Gain and Gini Index methods. Model evaluation employed AUC, Accuracy, and F1-Score metrics. The results demonstrate a significant improvement in Accuracy and F1-Score with the inclusion of preprocessing and feature selection stages, particularly for the Random Forest (RF) and Decision Tree (DT) algorithms. RF exhibited the highest accuracy of 97.1 %, indicating superior performance. Additionally, the AdaBoost (AB) and Gradient Boosting (GB) algorithms consistently demonstrate good performance, while the Neural Network (NN) algorithm shows substantial improvement when incorporating preprocessing and feature selection stages.